Inception has released Mercury 3, which it calls the most capable diffusion-based large language model to date. Unlike most models that generate text token by token, Mercury 3 uses text diffusion, and Inception frames the release as proof that this architecture can scale to compete with traditional autoregressive models on complex tasks.
The headline numbers are built around speed. The company says Mercury 3 runs at 1,068 tokens per second on NVIDIA GPUs — four to eight times faster than comparable models — with a one-million-token context window. Launch pricing is $0.10 per million input tokens and $0.35 per million output tokens, a 50% discount off list price. It also claims Mercury 3 outperforms GPT-6 Luna and Claude Haiku 5.5 across a range of benchmarks.
Early-access partners are already running it in latency-sensitive settings: browser-use agents, interactive education, real-time search agents, and customer support. In one test, Octen AI reported a median generation time of 0.9 seconds on smart glasses, while tutoring platform RevisionDojo said it could give students instant feedback “without giving up reasoning.”
Mercury 3 is available through the Inception API Platform, OpenRouter, Baseten, and Vercel, and is OpenAI API compatible, with 100 million free tokens for new accounts. Meanwhile, Mercury 2 and 2.5 are being sunset for new users.
Compare today’s fastest models in our ChatGPT, Claude, and Gemini profiles, or see our ChatGPT vs Gemini breakdown.
Source: Unite.AI
